SuffixArray* build(const DNASeqList& reads, size_t threads = 1) {
assert(!reads.empty());
size_t num_strings = reads.size();
// In the multiple strings case, we need a 2D bit array
// to hold the L/S types for the suffixes
char** type_array = new char*[num_strings];
for (size_t i = 0; i < num_strings; ++i) {
const DNASeq& read = reads[i];
size_t num_bytes = (read.seq.length() + 1) / 8 + 1;
type_array[i] = new char[num_bytes];
memset(type_array[i], 0, num_bytes);
}
// Classify each suffix as being L or S type
for (size_t i = 0; i < num_strings; ++i) {
const DNASeq& read = reads[i];
size_t len = read.seq.length() + 1;
// The empty suffix ($) for each string is defined to be S type
// and hence the next suffix must be L type
setBit(type_array, i, len - 1, 1);
if (!read.seq.empty()) {
setBit(type_array, i, len - 2, 0);
for (size_t j = len - 2; j > 0; --j) {
char curr = read.seq[j - 1], next = read.seq[j];
bool type = (curr < next || (curr == next && getBit(type_array, i, j) == 1));
setBit(type_array, i, j - 1, type);
}
}
}
// setup buckets
size_t bucket_counts[DNAAlphabet::ALL_SIZE];
size_t buckets[DNAAlphabet::ALL_SIZE];
// find the ends of the buckets
countBuckets(reads, bucket_counts, DNAAlphabet::ALL_SIZE);
//getBuckets(bucket_counts, buckets, DNAAlphabet::ALL_SIZE, true);
// Initialize the suffix array
size_t num_suffixes = std::accumulate(&bucket_counts[0], &bucket_counts[0] + DNAAlphabet::ALL_SIZE, (size_t)0);
LOG4CXX_DEBUG(logger, boost::format("initialize SA, strings: %d, suffixes: %d") % num_strings % num_suffixes);
SuffixArray* sa = new SuffixArray(num_strings, num_suffixes);
// Copy all the LMS substrings into the first n1 places in the SA
size_t n1 = 0;
for (size_t i = 0; i < num_strings; ++i) {
const DNASeq& read = reads[i];
for (size_t j = 0; j < read.seq.length() + 1; ++j) {
if (isLMS(type_array, i, j)) {
SuffixArray::Elem& ele = (*sa)[n1++];
ele.i = i;
ele.j = j;
}
}
}
// Call MKQS, first on the sequence and then on the index in the read table
LOG4CXX_DEBUG(logger, boost::format("calling mkqs on %d of %d suffixes(%f), using %d threads") % n1 % num_suffixes % ((double)n1 / num_suffixes) % threads);
{
SuffixRadixCmp radixcmp(reads);
SuffixIndexCmp indexcmp;
if (threads <= 1) {
mkqs2(&(*sa)[0], n1, 0, radixcmp, indexcmp);
} else {
mkqs2(&(*sa)[0], n1, 0, radixcmp, indexcmp);
}
}
LOG4CXX_DEBUG(logger, "mkqs finished");
// Induction sort the remaining suffixes
for (size_t i = n1; i < num_suffixes; ++i) {
(*sa)[i] = SuffixArray::Elem();
}
// Find the ends of the buckets
getBuckets(bucket_counts, buckets, DNAAlphabet::ALL_SIZE, true);
for (size_t i = n1; i > 0; --i) {
SuffixArray::Elem elem = (*sa)[i - 1];
(*sa)[i - 1] = SuffixArray::Elem(); // empty
const DNASeq& read = reads[elem.i];
char c = read.seq[elem.j];
(*sa)[--buckets[DNAAlphabet::torank(c)]] = elem;
}
induceSAl(reads, sa, type_array, bucket_counts, buckets, num_suffixes, DNAAlphabet::ALL_SIZE, false);
induceSAs(reads, sa, type_array, bucket_counts, buckets, num_suffixes, DNAAlphabet::ALL_SIZE, true);
// deallocate t array
for (size_t i = 0; i < num_strings; ++i) {
SAFE_DELETE_ARRAY(type_array[i]);
}
SAFE_DELETE_ARRAY(type_array);
return sa;
}
// Implementation of induced copying algorithm by
// Nong, Zhang, Chan
// Follows implementation given as an appendix to their 2008 paper
// '\0' is the sentinenl in this algorithm
void saca_induced_copying(SuffixArray* pSA, const ReadTable* pRT, int numThreads)
{
// In the multiple strings case, we need a 2D bit array
// to hold the L/S types for the suffixes
size_t num_strings = pRT->getCount();
char** type_array = new char*[num_strings];
for(size_t i = 0; i < num_strings; ++i)
{
size_t s_len = pRT->getReadLength(i) + 1;
size_t num_bytes = (s_len / 8) + 1;
type_array[i] = new char[num_bytes];
assert(type_array[i] != 0);
memset(type_array[i], 0, num_bytes);
}
// Classify each suffix as being L or S type
for(size_t i = 0; i < num_strings; ++i)
{
size_t s_len = pRT->getReadLength(i) + 1;
// The empty suffix ($) for each string is defined to be S type
// and hence the next suffix must be L type
setBit(type_array, i, s_len - 1, 1);
setBit(type_array, i, s_len - 2, 0);
for(int64_t j = s_len - 3; j >= 0; --j)
{
char curr_c = GET_CHAR(i, j);
char next_c = GET_CHAR(i, j + 1);
bool s_type = (curr_c < next_c || (curr_c == next_c && getBit(type_array, i, j + 1) == 1));
setBit(type_array, i, j, s_type);
}
}
// setup buckets
const int ALPHABET_SIZE = 5;
int64_t bucket_counts[ALPHABET_SIZE];
int64_t buckets[ALPHABET_SIZE];
// find the ends of the buckets
countBuckets(pRT, bucket_counts, ALPHABET_SIZE);
getBuckets(bucket_counts, buckets, ALPHABET_SIZE, true);
// Initialize the suffix array
size_t num_suffixes = buckets[ALPHABET_SIZE - 1];
pSA->initialize(num_suffixes, pRT->getCount());
// Copy all the LMS substrings into the first n1 places in the SA
size_t n1 = 0;
for(size_t i = 0; i < num_strings; ++i)
{
size_t s_len = pRT->getReadLength(i) + 1;
for(size_t j = 0; j < s_len; ++j)
{
if(isLMS(i,j))
pSA->set(n1++, SAElem(i, j));
}
}
/*
//induceSAl(pRT, pSA, type_array, bucket_counts, buckets, num_suffixes, ALPHABET_SIZE, false);
//induceSAs(pRT, pSA, type_array, bucket_counts, buckets, num_suffixes, ALPHABET_SIZE, true);
// Compact all the sorted substrings into the first portion of the SA
size_t n1 = 0;
for(size_t i = 0; i < num_suffixes; ++i)
{
SAElem elem = pSA->get(i);
if(!elem.isEmpty() && isLMS(elem.getID(), elem.getPos()))
{
pSA->set(n1++, elem);
}
}
*/
double ratio = (double)n1 / (double)num_suffixes;
std::cout << "[saca] calling mkqs on " << n1 << " suffixes " << ratio << " using " << numThreads << " threads \n";
// Call MKQS, first on the sequence and then on the index in the read table
SuffixCompareRadix radix_compare(pRT, 6);
SuffixCompareIndex index_compare;
//SuffixCompareID id_compare(pRT);
if(numThreads <= 1)
mkqs2(&pSA->m_data[0], n1, 0, radix_compare, index_compare);
else
parallel_mkqs(&pSA->m_data[0], n1, numThreads, radix_compare, index_compare);
std::cout << "[saca] mkqs finished\n";
// Induction sort the remaining suffixes
for(size_t i = n1; i < num_suffixes; ++i)
pSA->set(i, SAElem());
// Find the ends of the buckets
getBuckets(bucket_counts, buckets, ALPHABET_SIZE, true);
for(int64_t i = n1 - 1; i >= 0; --i)
{
SAElem elem_i = pSA->get(i);
pSA->set(i, SAElem()); // empty
char c = GET_CHAR(elem_i.getID(), elem_i.getPos());
pSA->set(--buckets[GET_BKT(c)], elem_i);
}
induceSAl(pRT, pSA, type_array, bucket_counts, buckets, num_suffixes, ALPHABET_SIZE, false);
induceSAs(pRT, pSA, type_array, bucket_counts, buckets, num_suffixes, ALPHABET_SIZE, true);
// deallocate t array
for(size_t i = 0; i < num_strings; ++i)
{
delete [] type_array[i];
}
delete [] type_array;
}
